In 2026, AI fluency is a baseline expectation for most knowledge work hires. The challenge is evaluating it honestly: candidates can fake high-level claims easily. Here are the interview questions and tasks that surface real fluency from someone who has actually built it.
Don't ask "do you use AI?" Almost everyone says yes. Instead: ask candidates to describe specific workflows they've built, walk through real prompts they use weekly, and complete a short paid take-home task using Claude or ChatGPT. The gap between surface familiarity and real fluency shows up immediately.
Most knowledge work hires in 2026 should be at level 3+. Leadership roles should be at level 4.
Paid 90-minute task: Give the candidate a real (de-identified) work artifact from your team: a customer email thread, a draft proposal, a meeting transcript. Ask them to produce a structured output (synthesis, draft, recommendation) using AI tools of their choice. They share their screen via Loom recording or share the artifacts.
Pay for the time ($100-$200 typical). The signal is enormous: you see how they actually work with AI, not how they describe it.
The same interview question produces very different answers depending on where a candidate actually sits on the fluency scale. Here is what hiring managers report hearing at each end of the spectrum.
High-fluency candidates default to specifics without prompting. Low-fluency candidates use the word "sometimes" heavily and rarely name a specific tool, prompt, or result.
Ask candidates to walk through a specific workflow they use weekly, show a prompt they have refined over multiple sessions, and describe a time AI gave them wrong output. These three questions separate genuine fluency from surface familiarity faster than any other approach.
The best method is a paid 90-minute take-home task: give candidates a real (de-identified) work artifact and ask them to produce structured output using AI tools of their choice, recorded via Loom. You see exactly how they work, not just what they claim.
For most knowledge work roles in 2026, yes. The minimum bar is Level 2 (casual weekly use). For leadership roles, Level 3 or 4 is the expectation: built personal workflows, refines prompts, and models AI use for their team.
Casual ChatGPT use means generic prompts for one-off tasks. Real AI fluency means built and iterated personal workflows, choosing between models based on the task, maintaining a prompt library, and having trained others on what works.
Hire if you have 90 days of onboarding bandwidth and the candidate demonstrates genuine willingness to learn. The real risk is the candidate who is defensive about AI. A coachable level-1 will outperform a resistant level-3 within a year.
Yes. For individual contributors, evaluate daily workflow integration and prompt quality. For executives, prioritize their track record rolling out AI to a team, their philosophy on AI augmentation vs. replacement, and whether they can model fluency for direct reports.
Partially. You can use Claude or ChatGPT to review the quality and structure of a candidate's take-home output. But the in-person conversation about how they built that output, what they iterated, and what did not work reveals the most about genuine capability.